[LAB 01 · 固定表示：标签预算与读出](/course/notebooks/4dd8605ed9904820919ea9e13d9e7d7c.html)

# 本地保存输出（非 Deepnote 云端运行）
[独立阅读](/course/notebook-blocks/0b4f1a8f114e48fd8c42a661c87b7303.html)
按问题继续

固定CNN与缓存特征、标签预算、kNN/linear和独立评价

**本地保存输出（非 Deepnote 云端运行）**

![本地保存图](/assets/display/d8ea2ed2099c1fdacd6f169c.webp)

```
Frozen features: (640, 32) ; all parameters unchanged.
method | labels | train | validation | heldout | heldout errors / 256
kNN 32 0.8438 0.7266 0.7148 73
linear 32 0.8438 0.6406 0.6133 99
kNN 64 0.9062 0.8828 0.9141 22
linear 64 0.8281 0.7422 0.6953 78
kNN 128 0.9375 0.9453 0.9453 14
linear 128 0.8359 0.7969 0.8125 48
kNN 256 0.9609 0.9609 0.9492 13
linear 256 0.8555 0.8203 0.8359 42

```